06-reference

every vibe check sonnet5

2026-07-02·reference·source: Every·by Katie Parrott (with Kieran Klaassen, Mike Taylor, Austin Tedesco)

Why this is in the vault

Every's collaborative Vibe Check finds that Sonnet 5 is competent across writing, coding, and knowledge work but fails to earn a clear default slot against any competitor — always losing to something faster, cheaper, or smarter. For an enterprise AI consultant choosing models for client demos and agent builds, this is a useful signal about where Sonnet 5 sits in the current stack.

Core argument

Mapping against Ray Data Co

Ray Data Co is a solo founder AI-native consultancy. The founder (Ray Wilson) is building AI agent infrastructure for enterprise clients (primarily through his phData DSA role as a Deal Solutions Architect), developing skills/plugins, and positioning as a Tier-5 AI operator. Relevant intersections: Anthropic model releases and their implications for agent builders, Claude's capabilities vs cost tradeoffs, what model to default to for production agent work.

Every's finding that Sonnet 5 underperforms on agentic builds is directly relevant to Ray's plugin and skill development — if the model stalls on multi-step agentic tasks, it's not a safe default for the Claude Code harness or for client-facing agent demos at phData. The pricing-doesn't-justify-it verdict from Mike Taylor matters for phData client conversations where Ray needs to recommend a model tier: Sonnet 5 may be hard to position as the "smart but affordable" middle option when Fable 5 or Opus 4.8 takes every head-to-head. For the Anthropic Claude Certified Architect cert prep, this review reinforces the importance of understanding model-tier tradeoffs as a testable domain — not just benchmark numbers but practical behavioral differences under real workloads. Ray's current production default (claude-sonnet-4-6 in this very session) may warrant re-evaluation as Fable 5 and Opus 4.8 mature in availability and cost curves.

Related